Use your own Anthropic or OpenAI key, and expose your brain to Claude, ChatGPT, or any MCP client — read-only and scoped. Your knowledge, portable everywhere.
Here's the usual bargain with an AI tool: the more value you build inside it, the harder it gets to leave. That's not an accident. It's the business model. Lock-in is a feature the vendor ships to itself at your expense.
I decided Zsper would take the opposite bet, and wire it into the architecture so it couldn't quietly erode later. Two features carry that bet: bring-your-own-key and MCP connectors. Together they mean your spend is transparent, your knowledge is portable, and your brain is usable inside whatever tool you already live in — including tools that aren't Zsper.
Prefer to run on your own model account? Paste in your Anthropic or OpenAI key and Zsper routes AI work through it instead of ours.
BYOK makes your spend transparent and portable. You're never locked into our pricing to keep using your own brain. If you've already negotiated rates or committed spend with a model provider, you keep them. And if you ever want to walk, your economics don't suddenly change on the way out — the knowledge you built is yours, and so is the account it runs on.
Vendor lock-in is a real line item, especially for a small team watching every recurring cost. The standard AI-tool trap is exactly the bargain I opened with: value in equals difficulty out. Zsper inverts it on purpose. Your brain is inspectable and exportable, your model account is your own, and the connector surface below means your knowledge isn't even trapped inside Zsper itself.
I'd rather earn the renewal every month than trap you into it. Those are different products, and they attract different founders.
There's a second, quieter benefit that matters more the bigger you get: cost transparency. On our metered credits, you see what each draft costs and we mark it up to run the business — fair, and honest about it. On BYOK, the markup disappears entirely; you pay your provider's rate and nothing else. For a team publishing at volume, that gap is real money, and you get to decide which side of it you want to sit on. Most tools would rather you never had that choice.
Setup takes about a minute. You paste your Anthropic or OpenAI key into Settings; it's encrypted with AES-256-GCM before it touches storage. From there, Zsper routes drafting, extraction, and embeddings through your account. You watch usage in the same dashboard you'd use for our metered credits — except these line items are informational, because your provider bills you directly. There's a fair-use cap to stop runaway abuse, but within it, BYOK usage is never charged by us. Remove the key and Zsper falls back to standard metered credits without missing a beat.
If you haven't met it yet: the Model Context Protocol (MCP) is an open standard that lets an AI assistant talk to external tools and data sources through one common interface. Instead of every app inventing its own plugin format, an MCP server exposes a small set of tools — like search and fetch — that any MCP-capable client can call. It's fast becoming the USB-C of AI integrations: one standard, many devices. Zsper speaks it, so your brain becomes one of those data sources.
Your knowledge shouldn't be trapped in one app, including mine. Zsper exposes each workspace's brain over MCP so any capable client can read it.
Mint a connector token in Settings and point any MCP-capable client at your endpoint:
Now your assistant can pull your real opinions, customer stories, and published articles into whatever you're doing — a cold email in Claude, a proposal in ChatGPT, a spec in your editor — grounded in what you actually think. Your positioning stops living only inside Zsper and starts showing up everywhere you work.
A few concrete examples from daily use:
In every case, the assistant isn't guessing what you think. It's retrieving it.
This is the part I'd underline. The MCP surface is read-only and workspace-scoped — not by policy, by design:
The tools it exposes are simple: search and fetch (the two ChatGPT connectors require), plus search_memory and search_articles for finer control. And the same grounding rule follows your knowledge everywhere it goes: use what the brain returns; never invent facts that aren't there.
Do I need BYOK to use the MCP connectors? No. They're independent. Run on our metered credits and still expose your brain over MCP, or bring your own key and never touch the connectors. Use either, both, or neither.
Can an MCP client accidentally change my brain? No. The surface is strictly read-only. External clients can search and fetch; they cannot create, edit, suppress, or delete a single record. Writing to your brain only ever happens inside Zsper, behind the same approval gates.
What happens if a token leaks? Revoke it in Settings and it stops working immediately. A token grants exactly one workspace and read-only access, so the blast radius is bounded even before you revoke — no cross-workspace reach, no ability to mutate anything.
Is my knowledge used to train anyone's model? No. Your records are yours; retrieval serves them to your own tools. Zsper doesn't train models on your writing, full stop.
Anthropic or OpenAI — which key should I bring? Either works, and you can pick on the models and rates you already have. Zsper's provider indirection means the choice is yours to change later without redoing anything; your brain and your drafts don't care which vendor sits underneath.
It would've been easy to lock all of this down — keep the keys ours, keep the knowledge inside our four walls, make leaving painful enough that you don't. Plenty of tools do exactly that. Zsper takes the opposite position on purpose, because an intelligence layer earns its place by being useful everywhere, not by holding your data hostage. BYOK keeps your economics portable. MCP keeps your knowledge portable. An asset you can't move isn't fully yours, and I'm not interested in selling you an asset with a lock on it.
The direction of travel across the whole industry is toward interoperability — models you can swap, context you can carry, tools that talk over open standards instead of proprietary walls. BYOK and MCP are Zsper's bet on that future landing. Rather than trying to be the one app you live in, Zsper aims to be the intelligence layer underneath the apps you already live in: the place your knowledge is captured and scored, then exposed cleanly to whatever assistant you're using this month.
That's a more durable position than a walled garden, and a better deal for you. Your knowledge outlives any single tool, including this one. As new MCP-capable clients appear, your brain plugs into them on day one, no migration required.
I'll admit the obvious tension: building the exit door is a strange thing for a founder to do. Conventional wisdom says make it hard to leave. But I think that logic is backwards for an intelligence layer. The tools that win the next decade won't be the ones that trapped the most data — they'll be the ones people kept choosing, month after month, because leaving was easy and staying was still the better call. If Zsper is only valuable when you can't escape it, it isn't actually valuable. So I'd rather compete on that footing, with the door propped open, than pretend a locked one is a feature.
Bring your own key, plug your brain into any LLM, and keep the option to walk away with everything you've built. That's the deal. Openness isn't a line in the marketing here — it's how the product is architected, all the way down.
Import what you’ve written, watch it learn, and ship a draft that sounds like you.
Start writing freeKeep reading
An AI writing tool with memory that remembers how you think — and builds every draft from what it knows about you, with receipts. Made for Indian founders and creators.
PillarThe pipeline is Knowledge → Understanding → Reasoning → Context → LLM — not plain RAG. Here's what happens before the AI writes a word, and why drafts sound like you.